When a new Last Mile test is created, the Synthetic Classic job queue server distributes test jobs to peers that match the test's defined peer population. If not enough peers are immediately available, the job queue server waits for more peers that match the peer population to check in, and will continue to assign these test jobs until enough data is received back so that the test's settings are honored.
The Synthetic Classic Portal uses this method to ensure that data is collected regardless of any errors that may occur in individual peer machines. However, this method may also result in some extra tests running. Since this process doesn't define a static number of tests, it is possible that a slightly different number of tests may have been collected when comparing any two time frames.
You can alleviate — but not eliminate — this issue by running your tests in a batch group. Combining tests into a batch group means that the tests will run on the same machines when the test is executed. However, test data is returned to the Synthetic Classic Portal in between test runs, so if a connection break occurs after the first test is done but before the second test is completed, you will see a discrepancy in the number of test runs.